Speed: 3D sensors, current amplifiers

1

Speed: 3D sensors, current amplifiersCinzia Da Via

Manchester University

Giovanni Anelli, Matthieu Despeisse1, Pierre Jarron

CERN

Christopher Kenney, Jasmine Hasi,

SLAC

Angela Kok

SINTEF

Sherwood Parker

University of Hawaii

Initial work and calculations with Julie Segal

Wall electrode data with Edith Walckiers, Philips Semiconductors AG

1. now at Ecole Polytechnique Fédérale de Lausanne (EPFL), Institute of Microengineering (IMT), Photovoltaics and thin film electronics laboratory, Neuchatel, Switzerland.

2

1. introduction

2. history

3. factors affecting speed

4. generating the signal – Ramo’s theorem

5. amplifying the signal – charge and current amplifiers

6. trench electrode sensors

7. hex-cell sensors

8. experimental results

9. analysis – constant fraction discrimination

10. analysis – fitting with almost-noise-free pulses

11. next

3

Potential 3D features from preliminary calculations by Julie Segal:

3. Fast pulses. Current to the p electrode and the other 3 n electrodes.

(The track is parallel to the electrodes through a cell center and a null point. V – bias = 10V. Cell centers are in center of any quadrant. Null points are located between pairs of n electrodes.)

1 ns 3 ns

50 µm

p

n

8 µm50 µm

4

1. shorter collection distance

2. higher average fields for any given maximum field (price: larger electrode capacitance)

3. 3D signals are concentrated in time as the track arrives

4. Landau fluctuations (delta ray ionization) arrive nearly simultaneously

5. drift time corrections can be made

1. 3D lateral cell size can be smaller than wafer thickness, so

2. in 3D, field lines end on electrodes of larger area, so

3. most of the signal is induced when the charge is close to the electrode, where the electrode solid angle is large, so planar signals are spread out in time as the charge arrives, and

4. Landau fluctuations along track arrive sequentially and may cause secondary peaks

5. if readout has inputs from both n+ and p+ electrodes,

Speed: planar 3D

4.

4.

4.

5

1. introduction

2. history





7. hex-cell sensors




11. next

6

A Very Brief History of Ever Shorter Times• The first silicon radiation sensors were rather slow with large, high

capacitance elements. The resultant noise was reduced by integration.

• For example, in the pioneering UA2 experiment at CERN, “the width of the shaped signal is 2 µs at half amplitude and 4 µs at the base.” (Faster discrete-component amplifiers were available, but not widely used.)

• The development of microstrip sensors greatly reduced the capacitance between the top and bottom electrodes, adding a smaller, but significant one between adjacent strips.

• The 128-channel, Microplex VLSI readout chip, had amplifiers with 20 – 25 ns rise times, set by the need to roll off amplification well before

• ω t ≤ π (t = time, input to inverted output then fed back to input)

• (Otherwise we would have produced a chip with 128 oscillators and no amplifiers.)

• The planned use of microstrip detector arrays at colliders with short inter-collision times required a further increase in speed.

• Silicon sensors with 3D electrodes penetrating through the silicon bulk allow charge from long tracks to be collected in a rapid, high-current burst.

• Advanced VLSI technology provides ever higher speed current amplifiers. Up to the sensor speed, such signals grow more rapidly with increasing frequency, than white noise.

7

The first ever custom VLSI silicon microstrip readout chips. Made at Stanford in 1984). (left, 7.5 cm), then by AMI – (right, 10 cm).

8

30 ns

planar sensor pulse shape

(an early, successful, attempt to increase speed in the era of 1

μs shaping times)

30 ns

9

1. introduction

2. history





7. hex-cell sensors




11. next

10

Some elements affecting time measurements

1. variations in track direction – 1 and 2 can affect the shape and timing of the detected pulse.

2. variations in track location3. variations in total ionization signal – can affect the trigger delay.

4. variations in ionization location along the track – Delta rays – high energy, but still generally non-relativistic, ionization (“knock-on”) electrons. Give an ever-larger signal when the Ramo weighting function increases as they approach a planar detector electrode, with their current signal dropping to zero as they are collected. This produces a pulse with a leading edge that has changes of slope which vary from event to event, limiting the accuracy of getting a specific time from a specific signal amplitude for the track.

5. magnetic field effects affecting charge collection – E × B forces shift the collection paths but for 3D-barrel only parallel to the track.

6. measurement errors due to noise – This currently is the major error source.

7. incomplete use of, or gathering of, available information – This is a challenge mainly for the data acquisition electronics which, for high speed, will often have to face power and heat removal limitations.

8. In addition, long collection paths for thick planar sensors increase the time needed for readout and decrease the rate capabilities of the system.

11

1. introduction

2. history





7. hex-cell sensors




11. next

12

1. Calculate E fields using a finite element calculation. (Not covered here.)

2. Calculate track charge deposition using Landau fluctuating value for (dE/dx) divided by 3.62 eV per hole-electron pair.

3. Paths of energetic delta rays may be generated using Casino, a program from scanning electron microscopy. (GEANT4 may be used for some of 2 and 3.)

4. Calculate velocities and diffusion using C. Jacoboni, et al. “A review of some charge transport properties of silicon” Solid-State Electronics, 20 (1977) 7749.

5. Charge motion will induce signals on all electrodes, each of which will affect all the other electrodes. Handle this potential mess with:

6. Next: charge motion, delta rays, Ramo’s theorem.

Calculating the signals

13

DELTA RAYS - 1

2

2

222 )(

2T

TFz

A

ZcmrN

dTdx

ndeeA

Integrating over T, the kinetic energy of the delta ray gives the number of delta rays in the 170 μm thickness of the hex sensor with T between T1 and T2

(Tmax is ≈ MeV; 1/Tmax ≈ 0)

21

1103.3

TTKeVn

So 3 KeV δ rays are common, 30 KeV uncommon, 300 KeV rare. Calculate production angles and then look at some of them.

2

222

max )/(/21

2

MmMm

cmT

ee

e

≈ MeV; 1/Tmax ≈ 0

14

• With electron velocities of about 5 x 106 cm / sec,

a delta ray of length 0.5 μm• if oriented ahead of the track• could reach an n electrode up to 10 ps ahead of

the main track.• This will happen above 10 KeV in ≈ 5-10% of

events• These energies will be compared with the mean loss

• dE/dxmin, silicon = 1664 KeV / gm / cm2 giving

• ΔTmean = 2.329 x 0.017 x 1664 = 65.9 KeV.

DELTA RAYS - 2

15

200 3-keV delta rays

0.1µm

16


1 µm

17


5 µm

18


15 µm

19

20

Velocities, diffusion, and collection times for a100 µm parallel-plate trench electrode gap.

electrons holes units

temperature 293.15 245* 293.15 245 °K

V (E = 0.5 V / µm) 4.93 7.0 2.07 2.22 cm/µs

t (E = 0.5 V / µm) 2.03 1.61 4.84 3.53 ns

σt, (parallel diffusion) 0.059 0.16 ns

V (E = 1.0 V / µm) 6.91 8.8 3.46 4.62 cm/µs

t (E = 1.0 V / µm) 1.45 1.21 2.89 2.22 ns

σt, (parallel diffusion) 0.029 0.06 ns

3 KeV δ ray (1 V / µm) 1.9 1.5 3.8 2.8 ps

10 KeV δ ray (1 V / µm) 14 11 29 22 ps

30 KeV δ ray (1 V / µm) 101 80 202 152 ps

60 KeV δ ray (1 V / µm) 362 284 723 541 psCalculations based on material in:

A REVIEW OF SOME CHARGE TRANSPORT PROPERTIES OF SILICON

Solid-State Electronics 20 (1977) 77 – 89

C. Jacoboni, C. Canali, G. Ottaviani and A. Alberigi Quaranta

21

1. introduction

2. history





7. hex-cell sensors




11. next

22

rise times ≈ 3.5 ns fall times ≈ 3.5 ns

0.13 µm chips now fabricated and used here

rise, fall times ≈ 1.5 ns

23

1. introduction

2. history





7. hex-cell sensors




11. next

24

next section offset so signal electrodes do not line up

signal electrodes with contact pads to readout

beam in

200 – 300 µm

active edge

Schematic diagram of part of one section of two of the planes in an active-edge 3D trench-electrode detector. Other offsets (⅓, ⅔, 0, ⅓,

⅔ ..etc.) may also be used.

25

A trench-electrode sensor will have:

• high average field / peak field,

• a uniform Ramo weighting field,

• an initial pulse time that is independent of the track position and,

• for two facing 100 μm gaps with a common electrode and a 250

μm thickness (in the track direction) a capacitance of 0.527 pF

per mm of height.

• For moderate to high bias voltage levels ( ~ 50 V ) and low dopant

levels ( ~ 5 ｘ 1011 / cm3 ) we can neglect V depletion ≈ 2 V, and

assume a constant charge-carrier drift velocity. After irradiation,

drift velocities will not be uniform, but will be faster as we raise

the bias voltage.

26

Schematic, idealized diagram of induced currents from tracks in a parallel-plate trench-electrode sensor.

Tracks ( ● ) are perpendicular, at the mid and quarter points.

Velocity (electrons) ≡ 3.0 × Velocity (holes).

time

Induced Current

electrons

holes 100 μm

n electrode

p electrode

27

28Talk title 28

SLAC MaskFE-I4 Sensor: 8

FE-I3 Sensors: 9

FE-I4 geometry test sensors: 55

Trench Sensors:12: 50 um n-2-p pitch, 1 mm long, 64 channels

5: 50 um n-2-p pitch, 5 mm long, 64 channels

23: 100 n-2-p pitch, 3 mm long, 16 channels

29

Fast Timing ExampleEffort was made to pinch-off and isolate both types of trenches by deleting the surfaces of p-spray. So both electron and hole signals can be readout separately.

30

Fast Timing Example

31

1. introduction

2. history





7. hex-cell sensors




11. next

32

But for now we used a 50 μm-side hex sensor (following slides)

1. with 20 V bias, at room temperature - 40V should be ok,

2. with each column of hexagons tied to a 0.13 μm current-amplifier channel (so large capacitance),

3. exposed to an uncollimated 90Sr beta source,

4. output to an oscilloscope triggered by the signal itself.

33

34

35

36

1. introduction

2. history





7. hex-cell sensors




11. next

37

-8

-6

-4

-2

0

2

4

-30 -20 -10 0 10 20 30

3d.speed.20v.09

triggeradjacentadjacent

time (ns)

a track in two and an induced pulse in the other (green) neighbor

38

-5

-4

-3

-2

-1

0

1

2

-30 -20 -10 0 10 20 30

3d.speed.20v.02


time (ns)

First, one problem with betas: an example of a possible angled track

distorting the pulse shape.(We will need real test beam data)

39

-10

-8

-6

-4

-2

0

2

-30 -20 -10 0 10 20 30

3d.speed.20v.01

triggeringadjacentadjacent

time (ns)

Uncollimated 90Sr betas, 20 C, hex sensor (20V bias) to 0.13 μm current amplifier, self-triggers,

event 1 of 99

30 ns

40

-10

-8

-6

-4

-2

0

2

-30 -20 -10 0 10 20 30

3d.20v.51

time (ns)

The middle event

41

-10

-8

-6

-4

-2

0

2

-30 -20 -10 0 10 20 30

3d.speed.20v.100


Uncollimated 90Sr betas, 20 C, hex sensor (20V bias) to 0.13 μm current amplifier, self-triggers,

event 99 of 99

30 ns

42

The single-column event with the largest expected timing error in the central scatter plot.

43

Pulse shape from the sum of the 6 largest pulses. τ-rise = 1.6 ns, fwhm = 2.90 ns. Note the trailing edge

hole current, and amplifier ringing.

30 ns

44

-10

-8

-6

-4

-2

0

2

-30 -20 -10 0 10 20 30

0.8 ns rise time pulse to cal. input

trigger channeladjacent channeladjacent channel

pu

lse

he

igh

t (m

V)

time (ns)

With a pulse from a pulse generator, with the 10% and 90% time points only 0.8 ns apart, we see an amplifier rise time of 1.5 ns. Sensor signals have rise times of 1.6 ns. So the

amplifier is currently the limiting element.

45

1. introduction

2. history





7. hex-cell sensors




11. next

46

Estimate the time resolution at room temperature with

• the hex sensor, and

•a preliminary version of a 0.13 µm integrated circuit readout

•using data from un-collimated 90-Sr βs (but only with tracks in the central channel).

•(A wall-electrode with parallel plates would give shorter times, but the hex sensor already has almost the same output rise time as a 0.8 ns input rise time pulse generator, so the output shape is primarily determined by the amplifier, not the sensor).

•To simulate a constant fraction discriminator set at 50% (where slope is steepest):

•Fit leading baseline, and measure noise,

•Fit top and find halfway point,

•ΔT = σ-noise / slope

•With wall-electrode sensor and a parallel beam,

• can do better fitting entire pulse.

σ-noiseΔT

47

Noise distribution from pre-pulse region with a Gaussian fit.

48

Noise distribution from pre-pulse with a Gaussian fit – log scale to show tails

sigma = 0.33166 +/- 0.0033 mVdirect standard deviation from the18,090 voltage values = 0.3218 mV

49

Fourier transform of noise: Gaussian, but not white

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

0 200 400 600 800 1000

Fourier transform of noise

rela

tiv

e s

ize

frequency (MHz)

50

0

5

10

15

20

25

2 4 6 8 10 12 14 16

number vs. amplitude

Co

un

ts

0

200

400

600

800

1000

1200

2 4 6 8 10 12 14 16

dt

(ps

)

Amplitude (mV)0 5 10 15 20

number vs. dt

Counts

σnoise

dt

Scatter plot of expected noise-induced timing errors, dt, vs. pulse amplitude, for 67 pulses and the projections of dt and amplitude distributions. σ (noise) = 0.33 mV.

51

1. introduction

2. history





7. hex-cell sensors




11. next

52

1. The approximately noise-free signal pulse shape from the digital oscilloscope shown earlier was scaled from small to large pulse heights.

2. A set of noise sequences was prepared from the pre-pulse base lines.

3. Each noise sequence was added, point-by-point, to the scaled-amplitude signal.

4. The scaled noise-free pulse was adjusted to have the same peak height as the pulse in 3, and the standard deviation of the point-by-point differences found as the relative timing of the two pulses was shifted point (62.5 ps) by point.

5. A parabola was fit to the minimum value of the standard deviation and the two values on each side, and its minimum used as the best-fit time.

6. This was repeated for each pulse height.

53

Expected time errors, dt, due to noise as a function of pulse height from the combined signal pulse shape added to 201 noise segments with dt determined from the standard deviation of time variation of the 50% point on the leading edge (Δ) and from the time variation of the best fit time of the combined signal pulse shape to the same shape plus noise (●). The mean value of the best fit times (○) is 24% of the fit values. The signal to noise ratio is 3 times the pulse height in mV.

54

1. introduction

2. history





7. hex-cell sensors




11. next

55

NEXT1. Reduce as far as practicable, the sensor capacitance.

2. Use reduced temperatures to reduce noise and double speeds.

3. Use an amplifier with the lowest possible noise, given the available space, heat removal capabilities, and speed requirements.

4. Use higher electric fields giving drift velocities ≈ saturation values.

5. Use trench-electrode sensors.

6. Use waveform recorders if a channel can fit within the area of a pixel. Only the large-amplitude part of the signal is needed. The baseline average can be kept as a single, updated number in storage.

7. Use multiple timing layers of detectors, if allowed by Coulomb scattering, space, and cost considerations – some possibly rotated to help with tracking,

8. Use a weighting factor, as suggested by the time-resolution vs. pulse height results, to favor layers having high signal-to-noise ratios.

9. Considering 6-8 above, use high-resolution position-tracking layers. The most accurate timing will be done by a system, not by one sensor – readout unit.

56

Some Partial Conclusions

• With the latest 3D results we have seen a decrease in pulse times by 3 orders of magnitude.

• There should be possibilities of silicon sensor systems with time resolution well below 100 ps.

• The lowest times will use some combination of multiple layers, lower capacitances, higher voltages than the 20V we used, 1/amplitude weighting, lower temperatures, and/or improved electronics.

• Improved, fast, compact, wave-form digitizers could help.

• We can expect generic electronics certainly will also be improved by industry.

Documents

Speed: 3D sensors, current amplifiers